Modelling of impact of water quality on recharging rate of storm water filter system using various kernel function based regression

In this paper, recharging rate of stormwater filter system is assessed by using predictive models of Gaussian Process (GP) and Support Vector Machines (SVM). Four kernel functions: normalized poly kernel, polynomial kernel, Pearson VII kernel (PUK) and radial basis kernel (RBF) were used with both modelling approaches (GP and SVM). A dataset consists of 678 measurements was collected from the experimental investigations on the infiltration of the storm-water filter system. Out of 678 observations, randomly selected 462 observations were used for training, whereas remaining 216 were used for testing the model. Input variables were comprise of cumulative time (T), the thickness of medium sand bed (B), size of medium sand (S) and concentration of impurities (Conc.), whereas the recharging rate (R) was considered as output. Correlation coefficient (C.C) and root mean square error (RMSE) were used to compare the performance of both modelling approaches. The evaluation of result suggests that Pearson VII based GP regression approach works well as compared to the other kernel functions based on GP and SVM models. Sensitivity analysis suggested that the size of medium sand (S) is an important parameter for predicting the recharging rate of stormwater filter system. Parametric study outcomes suggested that the higher size of medium sand increases the recharging rate, and a higher concentration of impurities in water decreases the recharging rate. Moreover, on expanding the thickness of the medium sand bed the recharging rate of the storm water filter system was observed to be increased.